I did a livestream of this as “D major thing.” Since then, I’ve minorly tweaked a few values to give slightly better results (in my opinion), and I added a pretty light show to the first page.
The premise of the patch is that there are four voices, composed of waveshaped triangle oscillators fed into filters. The voices are heard by cycling -1 to 1 sine waves (swung to a bit of a cross between a sine and a sawtooth); as the LFO swings the filter below audio ranges, the oscillators are silenced. A high-pass filter is used to make sure the sub-audio oscillators don’t become too loud. There are no VCAs (I removed them during the livestream).
At the same time, each of those sine wave LFOs’ rates are controlled by another sine wave LFO. These rate LFOs are passed through multipliers, changing their shape so that they have long troughs and quickly accelerating, short peaks. This causes the voices to speed up and slow down.
All of the LFOs are set to prime numbers for their rates, so that they don’t repeat the same sounds (very often; I haven’t done any calculations; I’m pretty confident it would take a real long time before the patch returned to its start condition).
Every few cycles, the LFOs change notes; this is determined probabilistically, so nothing is set in stone. The notes are chosen from switches; the select of the note is random, but the choice of notes is not. (If you would like to reprogram the patch, the switches are labeled as “note banks.”) The notes are chosen from D major, but since D major contains notes from other keys, it may modulate into them momentarily.
A looper listens to all of the voices and records them (this is based a little on probability and a little on paying attention to when all the voices’ LFOs rise above and fall below 0). This loop is then played back an octave up and in reverse.
All of the audio is then fed into a ping-pong delay and a plate reverb for ambience.